Feat: Docker environment for remote speech to text evaluation

[Epic-Eric]

Description

Docker creates a container where the operating system and dependencies are uniform and the setup process is streamlined.

To build the docker file, first change the directory to speech_to_text in the terminal from Simuleval parent folder:

cd examples/speech_to_text

Then, build the Docker image with:

docker build -t simuleval-speech-to-text:1.0 .

Next, run the remote evaluation server using the Docker image:

docker run -p 8888:8888 simuleval-speech-to-text:1.0

This binds port 8888 of the container (server) to port 8888 on the local machine (client).

To pass data to the server and execute remote evaluation, open another terminal, and change its directory to examples/speech_to_text. Finally, you can access the server with the following code for instance:

Example input

simuleval --remote-eval --remote-port 8888 \
    --source-segment-size 500 \
    --source source.txt --target reference/transcript.txt \
    --source-type speech --target-type text \
    --output output --quality-metrics WER

Example output

Type of change

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

How Has This Been Tested?

Tested locally first, as there were some bugs with the upstream repository as detailed here:
#109
Using COPY . /Simuleval instead of RUN git clone https://github.com/facebookresearch/SimulEval in the Dockerfile, I tested my local changes and ensured they worked as expected.

[xutaima]

Just curious why is nltk added here?

[Epic-Eric]

Not adding nltk made test_tree_pipeline_cmd in test_agent_pipeline.py to fail. It only happened recently, and wasn't an issue when I submitted the visualization PR. I have no clue why, but the error message told me that I need to nltk.download so I did